Forgotten Words: Benchmarking NeoBERT for Dementia Detection in Low-Resource Conversational Filipino and English Speech
This paper presents the first NLP-based dementia detection study for Filipino speech, constructing a parallel bilingual dataset of 4,000 DementiaBank-derived transcripts with manual Filipino translations. Five model families are evaluated across monolingual, zero-shot cross-lingual, and bilingual fine-tuning settings. English-trained BERT degrades sharply on Filipino (Macro-F1 = 0.455), but bilingual fine-tuning recovers performance to Macro-F1 = 0.969–0.973 across all transformer models. The key finding is that multilingual clinical NLP performance is driven by linguistic coverage during training rather than model scale or architecture.
Related guides (1)
Related events (8)
FilBench: Benchmarking LLM Capabilities in Filipino Language
FilBench is a new benchmark introduced to evaluate large language models on their ability to understand and generate Filipino. The benchmark targets a historically underrepresented language in NLP evaluation suites, assessing both comprehension and generation tasks. This work addresses gaps in multilingual LLM evaluation coverage, particularly for Southeast Asian languages.
Speech-based dementia screening using Whisper embeddings to compensate for nonverbal subtest omissions
Researchers present a speech-based evaluation system for the German Syndrom-Kurz-Test dementia screening battery, combining transcript-derived scores with Whisper embeddings to reduce transcription scoring errors. The system also approximates expert overall ratings even when motor (nonverbal) subtests are omitted, addressing a key accessibility limitation of speech-only assessment. Models show strong correlation with expert ratings and effective discrimination between cognitive status groups.
LLMs predict dementia and depression severity from clinical interview transcripts in zero-shot and feature-extraction settings
Researchers evaluate three open-weights LLMs (Mistral 3.1, DeepHermes, Qwen3) for predicting dementia and depression severity from speech transcripts of 154 German-speaking patients in standardized clinical interviews. The study introduces a new observer-based Global Depression Scale (GDS-D) and tests both zero-shot prediction and LLM-based feature extraction for Support Vector Regression. Zero-shot performs well for depression (MAE 0.60), while structured feature extraction reduces dementia assessment error by up to 35%; pause-enriched automatic transcripts match human transcription quality, suggesting viable fully-automated screening pipelines.
Sentence-Level Clinical Provenance Categorization for Multidisciplinary Hospital Summarization Using Fine-Tuned Llama-3
This pilot study presents a pipeline for categorizing sentence-level clinical provenance across multi-source hospital notes, targeting structured summarization in high-complexity settings like the NICU. The authors fine-tune Llama-3 8B and 70B models on MedSecId (MIMIC-III annotations), achieving Macro F1 above 92% in-domain. Cross-domain evaluation reveals a scale-dependent transfer effect: SFT substantially improves the 70B model (+7% Macro F1) but yields only marginal gains for the 8B model. A quantized fine-tuned 70B model outperforms its full-precision baseline while reducing compute, suggesting quantized adaptation is viable for structured clinical NLP tasks.
IndicBERT-HPA: Reliability-Oriented Multilingual Orthopedic Decision Support with Selective Verification Deferral
This paper presents a framework for classifying free-text orthopedic clinical notes in English, Hindi, and Punjabi, introducing IndicBERT-HPA, a domain-adaptive encoder augmented with language-aware orthopedic adapter heads. The system is evaluated against multilingual transformers, a DistilBERT baseline, and zero-shot LLMs, with zero-shot LLMs found substantially less effective than task-adapted encoders for closed-set clinical classification. IndicBERT-HPA achieves Macro-F1 of 0.8792 and AUPRC of 0.902 under natural clinical prevalence. A deterministic selective-verification layer combining confidence gating, evidence-consistency checking, and language-risk screening improves accuracy from 71.5% to 84.4% at 72.3% coverage on a 5,000-record held-out set.
Automated ICD Classification of Psychiatric Diagnoses Using NLP and LLMs
This study evaluates NLP and ML approaches for automating the mapping of free-text psychiatric descriptions to ICD diagnostic codes, using a dataset of 145,513 Spanish clinical records. Methods range from classical BoW/TF-IDF representations to transformer-based embeddings including e5_large, BioLORD, and Llama-3-8B. Fine-tuned e5_large achieved the best performance with a micro-F1 of 0.866, outperforming classical methods by capturing semantic nuance and medical terminology. The work highlights challenges of long-tail label distributions and ambiguity specific to psychiatric clinical language.
ENPMR-Bench: Benchmarking Proactive Memory Retrieval for Emotional Support Agents
This paper introduces ENPMR-Bench, a benchmark for evaluating Emotional Need-aware Proactive Memory Retrieval in memory-augmented language agents deployed for emotional support applications. The benchmark includes over 1,800 memory-augmented dialogues grounded in Maslow's hierarchy of needs, with structured mappings between emotional needs and supportive memory types. Experiments show that both embedding-based and LLM-driven retrieval paradigms fall significantly short of golden memory conditions on empathy scores, and while chain-of-thought prompting helps, a substantial performance gap remains. The work highlights a systematic gap in current agent memory systems when applied to affective rather than purely factual retrieval tasks.
BayLing-Duplex: Native full-duplex speech dialogue using a single autoregressive LLM
Researchers introduce BayLing-Duplex, a speech language model that achieves native full-duplex interaction — simultaneous listening and speaking — using a single autoregressive LLM with no auxiliary VAD or turn-taking module. Built by fine-tuning GLM-4-Voice on 400K samples plus a lightweight DPO stage, it reaches 92% turn-taking success and 100% interruption success on InstructS2S-Eval, and improves speech-response quality substantially over Moshi. The approach adds only special tokens to the standard vocabulary, making it portable across LLM architectures without architectural changes.